Development of Fire Scenarios for Car Parking...

Development of Fire Scenarios for Car Parking Buildings usingRisk Analysis

MOHD ZAHIRASRI MOHD TOHIR1 and MICHAEL SPEARPOINT1

1Department of Civil and Natural Resources EngineeringUniversity of CanterburyChristchurch 8140, New Zealand.

ABSTRACT

This paper describes a relatively simple probability quantitative risk analysis model to determine appropriatefire scenarios for car parking buildings. The approach introduces a dimensionless measurement defined as firerisk level by multiplying probability by consequence. For the development of fire scenarios for car parkingbuildings, the key variables for the fire risk analysis are identified as vehicle parking distribution probabilityand how vehicles then form clusters of neighbours, vehicle classification, vehicle fire involvement probability,and the severity of vehicle fires. The selection of clusters of neighbouring vehicles and whether all vehicles inthe cluster catch fire has the probability to affect the fire risk level. An example analysis is performed where asimple two-row, 100 space parking model with a 75 % vehicle occupancy and 0.90 tendency factor weightingis used to obtain the vehicle distribution probability combined with various data sourced from the literature.It is found from the example analysis that fire risk level is largely driven by the vehicle fire involvementprobability such that a single vehicle fire presents the worst case scenario in terms of fire risk.

KEYWORDS: risk, probablistic, car parking

INTRODUCTION

Vehicle parking buildings are commonly found in most modern urban environments. Such buildings can bestand-alone structures or attached to other occupancy types. The buildings can be multi-storey; above groundor below ground; be fully or partially enclosed; and be used to park a range of vehicle types (cars, vans, busesetc.). The usage characteristics of such buildings will depend on the service they provide: parking for patronsof a shopping mall, long-stay parking at an airport, parking for the residents of household units etc. Thisparticular research is focussed on car parking buildings rather than for other vehicle types such as trucks orbuses and the approach is similar to previous vehicle-fire related research [1], such that fire risk is equal toprobability multiplied by consequence.

There have recently been several significant vehicle fires in car parking buildings and in some cases thesehave led to fatalities. For example, in 2006 seven fire fighters were killed in a fire in an underground car parkin Gretchenbach, Switzerland [2]. Also in 2006 there was a car park fire in Bristol, United Kingdom where 22vehicles were destroyed in the incident and one person died in the occupancy above the car park [2]. In termsof design Roosefid and Zhao [3] state that there are standard fire scenarios for car parking buildings requiredby the French authorities. The scenarios are seven cars including a utility vehicle in the same parking row,four cars including a utility vehicle situated in two adjacent parking rows and one car located at any positionon the floor. These fire scenarios are applied so as to derive the most severe scenarios in terms of meeting fireresistance objectives. However, Roosefid and Zhao note that the greatest number of vehicles involved in a carparking fire was not more than three from incident statistics.

The life safety concerns of occupants and fire fighters and the appropriate design scenarios for structuraldesign have led to consideration of the impacts of fires in car parking buildings. There is the need for furtherresearch into how to determine reasonable fire scenarios and raises the possibility that a single set of scenariosmay not be applicable to all types of car parking buildings given the variations in design and use. The workpresented in this paper is part of a larger risk-based research project where the first step is to create designscenarios which will be used for subsequent analysis. These scenarios need to consider the relative number,layout and type of vehicles that could be present in a parking building; the likelihood that multiple vehiclescould burn simultaneously and the potential total energy that could be released by the burning vehicles. Thefocus in this paper is on a vehicle parking model that can identify the likelihood and magnitude of multiplevehicle clusters.FIRE SAFETY SCIENCE-PROCEEDINGS OF THE ELEVENTH INTERNATIONAL SYMPOSIUM pp. 944-957 COPYRIGHT © 2014 INTERNATIONAL ASSOCIATION FOR FIRE SAFETY SCIENCE/ DOI: 10.3801/IAFSS.FSS.11-944

944

Fire risk analysis is used to identify the impact of having a range of different vehicle fire scenarios in parkingbuildings. As a quantitative approach, the analysis establishes a dimensionless measurement for comparison,defined here as the fire risk level. For this research it is found that the probability component depends on anumber of factors which are explained in the remainder of the paper. The consequence component is definedas the severity of the fire and is represented by the vehicle peak rate of heat release, and this is also discussedfurther in the paper. Clearly the most severe fire scenario does not necessarily have the highest risk level asit is compensated by the likelihood of the scenario occurring. Essentially the question becomes: for a givenfire incident that starts in a specific vehicle what is the likely probability of a certain number of other vehiclesbeing parked in neighbouring spaces, what are the likely types of vehicles in those spaces in terms of theircombustible mass, will the fire spread to all of the neighbouring vehicles and what are the likely rate ofheat release available from each vehicle that will contribute to the total heat release? Then how likely is thisincident compared to the population of other similar incidents and which one of this population presents thegreatest fire risk level?

The objective of this paper is to present an approach to establish vehicle parking scenarios using a probabilis-tic quantitative risk analysis method by incorporating a relatively simple vehicle parking model, statisticaldata on vehicle fleets, measurements of passenger vehicle heat release and vehicle fire incident data. Theresulting risk analysis method could be used for the future specification of regulatory requirements for thedesign of car parking buildings but it has also been developed to be sufficiently flexible as such that it canbenefit designers and regulators for the assessment of specific car parking buildings.

FIRE RISK ANALYSIS

In order to perform the fire risk analysis, the first step is to be able to understand the day-to-day situation ina parking building and then list all the key variables that are potentially associated with vehicle fires in thebuilding. This approach follows the generalized concept for any fire risk analysis, i.e. to identify the hazardsand then to quantify consequence and probability of those hazards [4].

The key variables identified are the vehicle parking distribution probability, i.e. the probability of vehiclesbeing distributed in a particular pattern throughout the building at a given time; the vehicle classificationi.e. the composition of different vehicle types in a fleet; the vehicle fire involvement, i.e. the likely numberof vehicles involved in a fire using past incident data; and the severity of vehicle fires, where each of thesevariables is further explained in this paper. These variables are then used to create the necessary risk analysiscomponents and the combination of these component variables determines a specific vehicle parking firescenario.

Since the approach provides a numerical assessment, all of the key variables are quantitatively determinedfor each scenario. A probabilistic approach is used to demonstrate the severity of the fire as it relates to thelikelihood of a given vehicle population and classification. The fire risk level is obtained by multiplying vehi-cle parking probability, vehicles classification, vehicle fire involvement probability and vehicle fire severity.Thus, this approach is used as a basis of a comparison to determine which scenario provides the highest firerisk.

Since there are almost limitless parking configurations; numbers of parking spaces; and parking space ar-rangements the approach used here attempts to be as generic as possible. Scenarios provide a general resem-blance of the problem which can be related to most typical vehicle parking buildings. This generic approachis defines a simple two-row parking space arrangement as shown in Fig. 1 as a starting point for the research.For this approach, the number of parking spaces n can be up to any desired value and the number of vehiclesx can be up to n spaces. As an example, in Fig. 1, the value for parking spaces n is 12 and the number ofvehicles x is 5.

The potential for an open-ended level of depth for each component has led to the need to retain a consistentlevel of detail when deciding on how to obtain numerical inputs to the risk analysis. However even with thesomewhat simplified approach described here, the calculations applied in the fire risk analysis are automatedwith the creation of a parking simulation model using Visual Basic for Applications in Microsoft Excel.

FIRE SAFETY SCIENCE-PROCEEDINGS OF THE ELEVENTH INTERNATIONAL SYMPOSIUM pp. 944-957 COPYRIGHT © 2014 INTERNATIONAL ASSOCIATION FOR FIRE SAFETY SCIENCE/ DOI: 10.3801/IAFSS.FSS.11-944

945

1 3 5 7 9 11

2 4 6 8 10 12

1 2 3

4 5 3

n = 12 x = 5

Fig. 1. Generic scenario.

Vehicle parking probability

The vehicle parking probability is used to determine the relative location of parked vehicles at a given time,i.e. the distribution of a given number of vehicles, x across the available parking spaces, n. The distributionof vehicles is then used to identify clusters of neighbouring vehicles as discussed later. For this research theparking location distribution is managed by using a Monte Carlo approach. For example, considering the two-row model introduced in Fig 1, there are 12 parking spaces available for vehicle parking, however there areonly 5 vehicles to fill the spaces. Each Monte Carlo run distributes the 5 vehicles into the 12 spaces randomlyand therefore a particular scenario is formed. A successive application of the Monte Carlo method is thenused to construct the foundation for the vehicle parking distribution input to the risk analysis.

In reality the distribution of parked vehicles is influenced by human behaviour factors. The study of thesefactors in the search for a parking space is interesting field of study as the topic itself is very broad. Fromthe work by Waerden et al. [5], it is found that distance variables between parking spaces and other aspects(i.e. ticket machines, car park entrance, stairways and/or exit to final destinations) have an impact on parkingspace search behaviour. Thus a random approach to the Monte Carlo car placement is unlikely to resemblethe reality of the parking distribution. The car placement procedure has been modified to include a parkingtendency factor where it is assumed that vehicles tend to park at one end of the model to represent a distancevariable. This parking tendency factor is governed by a user-defined weighting which controls the probabilityof vehicles being parked at one end of the model. The parking domain is equally split into a pair of two-rowsections where a higher weighting results in a greater likelihood that a vehicle is placed in one of the pairover the other. As an example, this parking tendency factor can be visualized in Fig. 2 where a weighting of80 % is applied. In this example, the dotted lines represent the separation for the pair of two-row sections;Section 1 and Section 2 with Section 1 being nearer to the distance variable. A run of simulation will have 80% chance of a car to be randomly placed in Section 1 while there is 20 % chance of the car to be randomlyplaced in Section 2. This simplification, however, has its own limitations where the distributions within thesections are still random, thus if Section 1 is full, then vehicles in Section 2 will not be further affected by thedistance variable. This can be improved by dividing the two-row parking model into many smaller sectionsalthough this more complicated algorithm has not been implemented in this work.

The work by Waerden et al. [5] showed three sets of parking data at a specific parking building in the Nether-lands with a specific duration of time. These parking data consist of 4 two-row parking and 3 single-rowparking spaces as well as two distance variables; a railway station and a passenger exit to a shopping mall,but for this analysis only the two-row parking data were used to match with the simple model proposed inthis work. By assuming that the railway station is the dominant distance variable it can be inferred that theweighting of a tendency factor at peak times is around 0.90 while at off-peak times it is around 0.70.

The parameters necessary for this component of the parking simulation model is the number of parkingspaces, the number of vehicles, the number of iterations for the Monte Carlo simulations and the weightingfor tendency factor. The input range for these variables are virtually unlimited, however, the limitations of Mi-


946

1 3 5 7 9 11

2 4 6 8 10 12 n = 12

Distance Variable (e.g. Passenger exit)

Section 1 Section 2

Fig. 2. Generic car parking scenario for 12 spaces with tendency factor.

crosoft Excel restricts the input up to certain maximum values. The output from the Monte Carlo simulationsis the result for each iteration presented in an Excel spreadsheet for further analysis.

Vehicle classification probability

Since this research particularly focuses on car parking buildings, the scope of the study is limited to privateroad passenger vehicles. Current research by Tohir and Spearpoint [6] shows that there are numerous waysto categorise passenger vehicles and different jurisdictions have a variety of definitions for the purposes ofclassification. Some of the most common classifications are the vehicle engine size, the vehicle seating capac-ity, the vehicle dimensions (e.g. length, interior volume size), the vehicle curb weight, age, or wheelbase [7].For this work, the American National Standards Institute (ANSI) [8] classification of road passenger vehiclesbased on curb weight of the vehicle is adopted (Table 1) as the mass is identified as a key parameter relatedto the potential fire load of vehicles.

Table 1. ANSI classification of vehicle by curb weight.

Classification Curb weightPassenger car: Mini 1500 − 1999 lbs (680 − 906 kg)Passenger car: Light 2000 − 2499 lbs (907 − 1134 kg)Passenger car: Compact 2500 − 2999 lbs (1135 − 1360 kg)Passenger car: Medium 3000 − 3499 lbs (1361 − 1587 kg)Passenger car: Heavy ≥ 3500 lbs ( ≥ 1588 kg)Van / MPV Not definedSUV Not defined

Following on from the selection of an appropriate vehicle classification system, associated statistics for theproportion of the road passenger vehicle types are presented to the fire risk level calculation. The proportionstatistics are used by the parking model to select the classification of a vehicle applied to a simulation. Thestatistics for composition of this classification is obtained from data from the USA [9] and from the EuropeanUnion [10] and is shown in Table 2.

Vehicle fire involvement probability

This component uses statistics from past vehicle fire incidents in car parking buildings as input into the firerisk analysis. Incident statistics are typically obtained from organizations that provide emergency fire fightingand rescue services where the nature of the details available depends the particular individual organization. InNew Zealand these statistics are extracted from the New Zealand Fire Services (NZFS) fire incident reportingFIRE SAFETY SCIENCE-PROCEEDINGS OF THE ELEVENTH INTERNATIONAL SYMPOSIUM pp. 944-957 COPYRIGHT © 2014 INTERNATIONAL ASSOCIATION FOR FIRE SAFETY SCIENCE/ DOI: 10.3801/IAFSS.FSS.11-944

947

Table 2. Composition of vehicle classification.

Classification Percentage compositionPassenger car: Mini 7%Passenger car: Light 16%Passenger car: Compact 20%Passenger car: Medium 20%Passenger car: Heavy 11%Van / MPV 10%SUV 16%

system (FIRS). For vehicle-related fires FIRS contains records for the date and time of incident, the incidenttype, the number of vehicles involved, the vehicle types, the vehicle year of manufacture, general propertyuse, specific property use, location of origin, heat source, objects ignited, and fire cause.

However, these statistics often do not provide a high level of detail regarding the incident. For example itcan be difficult to determine at what stage of the fire NZFS intervention took place or whether an automaticsprinkler system activated to suppress the fire. The statistics also do not state the total number of vehicles inthe car park at the time of the fire or the relative parking space locations of the fire-affected cars and thosenot affected. It can therefore be hard to know whether a vehicle fire in a parking building had the potential tospread to other vehicles had it been allowed to continue unchecked.

Earlier research by Li and Spearpoint [11] shows that the probability of a vehicle catching fire in car parkingbuildings in New Zealand from 1995 - 2003 was 4.74 × 10-6 per year. In this paper the analysis was extendedusing the same approach used by Li and Spearpoint up until 2012 using data from 2004 - 2012 obtained fromthe NZFS [12]. The probability for 2004 - 2012 was 1.15 × 10-6 per year, which is lower than the previousresearch making the overall probability from 1995 - 2012 as 2.76 × 10-6 per year. This probability is coupledwith the vehicle fire involvement statistics to produce a vehicle fire involvement probability.

Vehicle fire involvement statistics have been obtained from the reported fire incidents in car park buildingsacquired from the NZFS [12]. These statistics were strengthened by the collection of fire incident statistics incar park buildings compiled by Joyeux et al. [13] in 2002. The combined fire incident statistics is shown inTable 3. The table also shows the vehicle incidents probability and the annual vehicle fire involvement proba-bility where the vehicle incidents probability is the number of incidents for a particular cluster divided by thetotal vehicle fire incidents and the vehicle involvement probability is the probability of a vehicle catching firecoupled with the vehicle incidents probability. Also from Table 3, there were a total of 401 incidents reportedand the greatest number of vehicles involved were 7 with two incidents. The highest number of fire incidentsare single vehicle cases with 344 incidents.

Table 3. Numbers of vehicles involved in fire and number of fire incidents.

Number of vehicles involved Number of incidents Probability of incidents Vehicle fire involve-ment probabil-ity per year

1 344 0.858 2.37 × 10-10

2 27 0.067 1.86 × 10-11

3 21 0.052 1.45 × 10-11

4 4 0.010 2.75 × 10-12

5 3 0.007 2.06 × 10-12

6 0 0.000 No data7 2 0.005 1.38 × 10-12

Since the fire risk analysis requires data for up to maximum occupancy number of vehicles and the number ofincidents only involves a maximum of 7 vehicles, a correlation for vehicle fire involvements against numberof vehicles has been made. This correlation is used to predict the probability of a fire scenario occurring forFIRE SAFETY SCIENCE-PROCEEDINGS OF THE ELEVENTH INTERNATIONAL SYMPOSIUM pp. 944-957 COPYRIGHT © 2014 INTERNATIONAL ASSOCIATION FOR FIRE SAFETY SCIENCE/ DOI: 10.3801/IAFSS.FSS.11-944

948

higher numbers of vehicles than can be determined from the statistics. For this purpose, a simple correlationis obtained and shown in Fig. 3. A power law fit is used to correlate the known data because from the limitedobservation it is expected that the probability of incidents involving more vehicles will reduce. From thiscorrelation an equation of y = 0.66×10-6 x-2.67 where x is the number of vehicles and y is the probability ofincidents is obtained. Thus, this equation is used to predict the vehicle fire incident probability for more than7 vehicles.

y = 0.66x-2.67 R² = 0.96

0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

0 1 2 3 4 5 6 7

Pro

bab

ility

of

inci

de

nts

Number of vehicles

Fig. 3. Correlation of probability of incidents over the number of vehicles involved.

Consequence

For the consequence component of the risk analysis the heat release rate of road passenger vehicles is taken tobe the critical parameter in that a higher heat release rate contributes to a higher fire risk level. The aforemen-tioned research by Tohir and Spearpoint presents a distribution analysis for the fire severity characteristics ofsingle passenger road vehicles using published heat release rate data [6]. The work collates full-scale labo-ratory experiment data from 41 single passenger road vehicles in the form of the peak rate of heat release,the time to reach peak rate of heat release and total heat released. Even though in that work only four classeswere analysed i.e. Passenger Car: Mini, Light, Compact and Medium; the remaining classes can be estimatedthrough the frequency data plot of the vehicle peak heat release rate against the vehicle curb weight.

Fig. 4 shows an example of the distribution plot of peak heat release for Passenger Car: Mini classificationwith the 5th and 95th percentile values indicated. A best-fit Weibull distribution has been determined for thedata from 6 individual vehicle fire experiments. Average values for peak heat release rate are calculated foreach classification that are then used for this study. The average values, 5th and 95th percentile distributioncharacteristics for each classification are shown in Table 4. However, due to limited data sets in the previouslymentioned research work by Tohir and Spearpoint, the distribution characteristics for Passenger Car: Heavy,Van/MPV and SUV vehicle classification are extrapolated from the lower curb weight classifications.

It is noted that the procedures, standards and/or protocols varied between each experiment which likely leadto different effects on the fire spread, availability of air etc. and that the various heat release rate measurementtechniques, namely mass loss rate, convective calorimetry and species-based calorimetry, could result invariability in the heat release rate measurements. However, due to limited data sets in each curb weight


949

1,710 4,549

0.00E+00

1.00E-02

2.00E-02

3.00E-02

4.00E-02

5.00E-02

6.00E-02

7.00E-02

8.00E-02

9.00E-02

0

1000

2000

3000

4000

5000

6000

Pro

bab

ility

de

nsi

ty

Peak heat release (kW)

Weibull

5th percentile 95th percentile

Fig. 4. Distribution plot of the vehicle peak heat release rate for Passenger Car: Mini.

Table 4. Average estimated peak heat release rate values with its distribution characteristics foreach classification.

Classification Average, kW 5th percentile, kW 95th percentile, kWPassenger car: Mini 3490 1710 4549Passenger car: Light 4550 846 9802Passenger car: Compact 4150 1352 7406Passenger car: Medium 6890 3009 10850Passenger car: Heavy 8000 1849 13705Van / MPV 7000 1604 12016SUV 7000 1604 12016

classification group meant it was not possible to create absolutely homogenous data sets that also providesufficient items of data to be meaningful.

APPLICATION OF THE RISK APPROACH

Cluster size assessment

An example of the approach can be illustrated by presenting a simple parking problem. A single row ofparking spaces is used for easier understanding of the process where the case of 5 parking spaces with 3vehicles is illustrated. Fig. 5 shows all of the possible parking distribution scenarios for this case.

Two methods to determine the number of possible fire scenarios are described. For both methods the assump-tions made are:

• Only full vehicle fire involvement is considered; either a vehicle has caught fire or it has not, there isno partial vehicle fire.


950

(a) (d) (g)

(b) (e) (h)

(c) (f) (i) Fig. 5. All distribution scenarios for 5 parking spaces with 3 vehicles.

• There is no time dimension in the fire risk analysis, i.e. fires occur instantaneously and simultaneously.

• For each vehicle on fire, a peak heat release rate is selected to maximise the risk.

• Fire spread does not occur across gaps formed by empty parking spaces.

The two methods are as follows:

i) Method 1 - In this method, vehicles located in contiguous parking spaces are considered to be a clus-ter such that they all catch fire simultaneously. Thus using Figure 5 distribution scenario (a), showsa cluster of three vehicles which means only a single fire scenario is present. However Figure 5distribution scenario (b), shows a cluster of two vehicles parked next to each other and a separatecluster which consists of a single vehicle. Thus in distribution scenario (b) there are two clusters andtherefore two possible fire scenarios.

ii) Method 2 - This method is an expansion on Method 1 in which each cluster can have all of theconceivable fire scenarios available to represent the possibility that the fire does not spread to neigh-bouring vehicles regardless of the size of fire. For example, in Figure 5 distribution scenario (a),there is one case of 3 vehicles catching fire, two cases of 2 vehicles catching fire and three casesof 1 vehicle catching fire. Therefore in total there are six possible fire scenarios within the singledistribution scenario. For Figure 5 distribution scenario (b), there are two separate clusters but interms of probable fire scenarios, there is one case of 2 vehicles catching fire and three cases of asingle vehicle fire.

The calculation of the total possible number of fire scenarios and the associated probabilities for the exampleusing the two methods is shown in Table 5. It is evident that the probability of one vehicle on fire usingMethod 1 is 0.50 whereas for Method 2 it is 0.66 which is a 16 % difference. For two vehicles there is a 4% difference and for 3 vehicles there is a 12 % difference. These differences in probability show that usingalternative assumptions for the possibility of fire occurring in multiple vehicles will result different outcomesin the fire risk level. For this research only Method 1 is adopted and further studied in detail although thepossible implications of Method 2 are discussed.

Maximum occupancy

The starting point to determine the vehicle parking probability is to obtain data for parking occupancy valuesat different times and for car parking buildings that exhibit different usage characteristics. These data willFIRE SAFETY SCIENCE-PROCEEDINGS OF THE ELEVENTH INTERNATIONAL SYMPOSIUM pp. 944-957 COPYRIGHT © 2014 INTERNATIONAL ASSOCIATION FOR FIRE SAFETY SCIENCE/ DOI: 10.3801/IAFSS.FSS.11-944

951

Table 5. Total possible fire scenarios and probability for both methods.

Method 1 Method 2Number of vehicles Frequency Probability Frequency Probability

1 8 0.50 27 0.662 5 0.31 11 0.273 3 0.19 3 0.07

Total 16 1.00 41 1.00

present the parking trends for specific parking buildings. For this research, online data from several carparking buildings in San Francisco and Santa Monica, USA and two airport parking buildings in Switzerlandand Italy have been obtained.

An example of this data is taken from the Sutter-Stockton Garage in San Francisco. This 24 hour car parkprovides parking spaces for nearby shops and offices. The parking data for this particular building is taken fora typical week from 5th of May 2012 until 5th of May 2013. Fig. 6 shows the normalized parking space oc-cupancy as the number of vehicles parked over the total number of spaces available. Therefore the maximumoccupancy is on Thursday where it almost reaches 75 %. This maximum occupancy provides a measure ofthe maximum exposed fire risk and thus a value of 75 % is taken as starting point for the parking simulationmodel.

Saturday Sunday Monday Tuesday Wednesday Thursday Friday

12:00 PM 12:00 PM 12:00 PM 12:00 PM 12:00 PM 12:00 PM 12:00 PM 12:00 AM 12:00 AM 12:00 AM 12:00 AM 12:00 AM 12:00 AM

0

10

20

30

40

50

60

70

80

Pe

rce

nta

ge, %

Time and days of the week

Fig. 6. Sutter-Stockton parking garage distribution in different days of the week.

Accumulated peak rate of heat release

For each simulation run, the model specifies the location of each vehicle in the parking area and the class ofthe vehicle, from which the peak heat release rate of each vehicle can be identified. A single iteration of thesimulation will select vehicle classes based on the vehicle classification probability distribution. Thus, everysingle iteration will produce a different accumulated peak heat release rate. By executing a sufficiently largeFIRE SAFETY SCIENCE-PROCEEDINGS OF THE ELEVENTH INTERNATIONAL SYMPOSIUM pp. 944-957 COPYRIGHT © 2014 INTERNATIONAL ASSOCIATION FOR FIRE SAFETY SCIENCE/ DOI: 10.3801/IAFSS.FSS.11-944

952

number of iterations, a range of possible scenarios is obtained. To simplify the analysis, these ranges of peakheat release rate for a given number of vehicles are averaged.

The accumulated peak heat release rate for vehicles are recorded from each simulation iteration. Trial runsof 10,000 iterations for single vehicle fire up to 11 simultaneous vehicle fires are recorded. This is shown inFig. 7, where it can be seen that the average peak heat release rate for a single vehicle to 11 vehicles showsa linear fit. In Fig. 7 also shows the range of total accumulated peak heat release rate from the iterations. Toverify the linearity assumption, 10,000 iterations is run for the peak heat release rate accumulation for 20vehicles. Thus, the equation of the linear fit is used in the fire risk analysis to obtain the peak heat release ratefor a specified number of vehicles.

It is also noted that the usage of probability distribution in the simulation will produces outlier(s) based onthe extreme ends of the distribution shape. This explains why, for example for a single vehicle fire can reachover 40,000 kW.

y = 5961x

0

20000

40000

60000

80000

100000

120000

140000

0 2 4 6 8 10 12 14 16 18 20

Pe

ak h

eat

re

leas

e r

ate

, kW

Vehicle cluster size

Fig. 7. The total peak heat release rate for increasing vehicle cluster size.

EXAMPLE ANALYSIS

An example of the application of the fire risk analysis approach is demonstrated with the inputs being 100parking spaces, 75 vehicles (i.e. a 75 % occupancy) and 10,000 iterations for the parking simulation model.A two-row parking arrangement with a tendency factor weighting of 0.90 and Method 1 for determining firescenarios is used in this example.

The outcome of the parking simulation model is in the form of the probability of having different clustersizes, the vehicle involvement probabilities and total rates of heat release. The probability of having differentcluster sizes is shown in the second column of Table 6. However, the results are not shown for all 75 vehicleclusters because not every one is obtained through the 10,000 iterations. Thus, only clusters of vehicles withprobability results are shown where it can be seen that the highest occurs for the 51-vehicle cluster at 0.174.The second highest probability is the cluster with 52 vehicles at 0.135. This means that during the iterations,


953

the simulation model tends to repeatedly form the 51 or 52 vehicle clusters in preference to other clustersizes.

The third column in Table 6 is the vehicle involvement probability obtained using Fig 3 and the total rate ofheat release for a given cluster size is obtained from Fig. 7. The total heat release rate shows values that canexceed 380 MW which is a manifestation of the assumption that all cars ignite and burn simultaneously. How-ever whether this value could be achieved would also depend on the ventilation available within a particularcar park and any modelling would need to account for such conditions.

The fire risk level in Table 6 shows that the highest risk of vehicle fire is for a single vehicle at 4.90 ×10-4. Even though, the total accumulated heat release rate for a single vehicle is low, the vehicle involvementprobability governs the whole fire risk level. This is due to the large difference in the orders of magnitudesince the vehicle involvement probability follows a power law.

Table 6. Simulation and fire risk analysis by using Method 1.

Number ofvehicles

Cluster size probabil-ity

Vehicle involvementprobability

Total rate of heat re-lease (kW)

Fire risk level

1 0.041 2.00× 10-6 5952 4.90× 10-4

2 0.032 3.61× 10-7 11913 1.36× 10-4

3 0.042 1.32× 10-7 17874 9.94× 10-5

4 0.038 6.51× 10-8 23835 5.94× 10-5

5 0.033 3.75× 10-8 29796 3.65× 10-5

6 0.029 2.39× 10-8 35757 2.50× 10-5

7 0.020 1.63× 10-8 41718 1.36× 10-5

8 0.016 1.17× 10-8 47680 8.95× 10-6

9 0.013 8.77× 10-9 53641 6.21× 10-6

10 0.007 6.76× 10-9 59602 2.79× 10-6

11 0.006 5.34× 10-9 65563 2.26× 10-6

12 0.005 4.31× 10-9 71524 1.48× 10-6

13 0.003 3.54× 10-9 77485 7.60× 10-7

14 0.002 2.94× 10-9 83446 5.50× 10-7

15 0.001 2.48× 10-9 89407 1.78× 10-7

18 0.001 1.58× 10-9 107291 8.15× 10-8

20 0.001 1.22× 10-9 119213 7.75× 10-8

48 0.001 1.40× 10-10 286124 5.13× 10-8

49 0.004 1.33× 10-10 292085 1.53× 10-7

50 0.027 1.27× 10-10 298046 1.01× 10-6

51 0.174 1.21× 10-10 304007 6.39× 10-6

52 0.135 1.15× 10-10 309968 4.80× 10-6

53 0.099 1.10× 10-10 315929 3.43× 10-6

54 0.071 1.05× 10-10 321890 2.38× 10-6

55 0.059 1.00× 10-10 327851 1.93× 10-6

56 0.043 9.58× 10-11 333812 1.38× 10-6

57 0.029 9.17× 10-11 339773 9.00× 10-7

58 0.023 8.78× 10-11 345735 7.04× 10-7

59 0.016 8.42× 10-11 351696 4.66× 10-7

60 0.013 8.08× 10-11 357657 3.70× 10-7

61 0.007 7.75× 10-11 363618 1.83× 10-7

62 0.005 7.45× 10-11 369579 1.37× 10-7

63 0.005 7.16× 10-11 375540 1.36× 10-7

64 0.002 6.89× 10-11 381501 4.48× 10-8


954

A sensitivity analysis has been carried out by varying the weighting of the tendency factor in the exampleanalysis. In this analysis, the same number of parking spaces and number of vehicles were used while thetendency factor weighting is differed from 0.70 to 0.90 based on the analysis of Waerden et al.’s [5] data.

Fig. 8 shows how the parking tendency factor alters the average parking probability of each multiple vehiclecluster size. It is obvious that as the tendency factor weighting increases it will produce greater probabilitiesof large vehicle clusters. This sensitivity analysis also considers a random distribution i.e. a tendency factorweighting of 50 % which is shown by the × symbols. The addition of the random distribution is presentedfor the purpose of comparison as people invariably have a range of parking behaviour tendencies [14] thatwould mean it is not a random process. However it is interesting to note that when the distribution is randomit produces the highest probability of a single vehicle cases.

Fig. 9 compares the fire risk level for different tendency factor weightings i.e. 0.70, 0.80 and 0.90, and therandom distribution. The graph is shown using a semi-log scale to more clearly illustrate the wide range inthe results as the cluster size increases. From Fig. 9, varying the tendency factor weighting also affects the firerisk level even though it does not change the fact that a single vehicle fire has the highest fire risk level. Thisshows the importance of the vehicle involvement probability over the variations in the cluster size probability.The random distribution shows the highest fire risk level for a single vehicle due to the cluster size probabilitybeing directly related to the fire risk level.

0.00

0.02

0.04

0.06

0.08

0.10

0.12

0.14

0.16

0.18

0 10 20 30 40 50 60 70

Pro

bab

ility


0.90

0.85

0.80

0.75

0.70

Random

Weighting

Fig. 8. Cluster size probabilities for 75 vehicles in 100 parking spaces with different parking tendency factorweightings.

CONCLUSION

There are several limitations upon the fire risk analysis method used in this work. Firstly the online datafor parking probability are limited by the range of parking building data available and the distribution ofthe parked vehicles across the spaces is not included. This limitation could be addressed by making on-siteobservations in the required car parking building. Secondly the vehicle fire involvement probability usedstatistics that were a combination of data from different agencies and years. Finally the consequence part waslimited due to inadequate rate of heat release data for vehicle experiments that cover Van/MPV, SUV andPassenger car: Heavy classifications.


955

1.00E-08

1.00E-07

1.00E-06

1.00E-05

1.00E-04

1.00E-03

1.00E-02

0 10 20 30 40 50 60 70

Fire

ris

k le

vel


Tendency factor weighting 0.90



Random

Fig. 9. Fire risk level for 75 vehicles in 100 parking spaces with different parking tendency factorweightings.

By using Method 1 to find the fire scenarios, the highest fire risk is for a single vehicle at 4.90 × 10-4 for a75 % occupancy. More vehicles involved means higher consequences but the vehicle involvement probabilitygoverns the whole fire risk analysis since it shows significant difference in the order of magnitude of theprobability. Thus, more attention to the collection of vehicle involvement probability is needed in futurestudies. The next steps in this research are to examine the fire growth characteristics of car fires rather than toonly consider the peak rate of heat release and to model the spread of fire between cars using a tool such asB-RISK [15].

It is also noted that the current data for the vehicle involvement probability does not mention whether anysuppression systems were operated or at what stage any fire fighters intervention occurred. Had the informa-tion regarding the suppression of the fire in the statistics been included, a more realistic analysis is likely tobe produced. Furthermore the statistics do not indicate whether there were neighbouring vehicles present inthe incident which could have got involved in the fire. These limitations in the statistics have an impact onthe ability to provide appropriate data for a risk analysis model.

An initial assessment of Method 2 to find the fire scenarios suggests that it is likely to produce highest riskfor a single vehicle due to a greater weight of probability of having a single vehicle fire. It could be arguedthat the formation of scenarios using Method 2 already incorporates the vehicle involvement probability. Thissets grounds for more research to be carried out in the future.

The flexibility of the model allows for future analysis of car parking buildings with different number ofspaces, different occupancy numbers and the effect of human vehicle parking behaviour. In trying to achievethe objective of this research it is acknowledged that there is a continued interest in the phenomenon oftravelling fires in which a fire in a large space only burns over a limited area at any one time [16]. A carparking building is identified as one type of structure with the potential for travelling fires. However the firerisk analysis approach discussed here does not try to incorporate travelling fires as it requires more work tobe done should it be desirable to include this phenomenon.


956

REFERENCES

[1] Cheong, M.K., Spearpoint, M., and Fleischmann, C., (2008) Using Peak Heat Release Rate toDetermine the Fire Risk Level of Road Tunnels, Journal of Risk and Reliability 222: 595-604,http://dx.doi.org/10.1243/1748006XJRR169

[2] Shipp, M., “Fire spread in car parks,”Technical Report BD2552, Building Research Establishment(BRE), London,United Kingdom, 2010.

[3] Roosefid, M., and Zhao, B., “Fire safety engineering of an open carpark. Example use of ISO TS24679,”CTICM, Paris, France, 2011.

[4] Watts, J.M., and Hall, J.R.. “Introduction to Fire Risk Analysis,” The SFPE Handbook of FireProtection Engineering (4rd ed), DiNenno P.J. (ed.), National Fire Protection Association, 2008.

[5] Waerden, P.V.D., Borgers, A., Timmermans, H., “Travelers Micro-Behavior at Parking Lots: AModel of Parking Choice Behavior,” Processing of the 82nd Annual Meeting of the TransportationResearch Board, Washington DC, 2007.

[6] Mohd Tohir, M.Z., and Spearpoint, M., (2013) Distribution analysis of the fire severity charac-teristics of single passenger road vehicles using heat release rate data, Fire Science Reviews 2:5,http://dx.doi.org/10.1186/2193-0414-2-5

[7] Opland, L., “Size classification of passenger cars,” Masters Thesis, Chalmers University of Tech-nology, Goteborg, Sweden, 2007.

[8] American National Standard. “Manual on classification of motor vehicle traffic accidents, ” ANSID16.1-2007, 2007.

[9] Subramaniam, R., “Passenger vehicle occupant fatality rates by type and size of vehicle,”DOT HS809 979. Traffic Safety Facts. Research Note, January 2006.

[10] European Union. “Regional transport statistics, ”Eurostat: Regional Yearbook 2012, 2012.

[11] Li, Y., and Spearpoint, M., (2007) Analysis of vehicle fire statistics in New Zealand parking build-ings, Fire technology, 43(2): 93 - 106, http://dx.doi.org/10.1007/s10694-006-0004-2

[12] Challands, N., Personal communication, 2012.

[13] Joyeux, D., Kruppa, J., Cajot, L.-G., Schleich, J.-B., de Leur, P.V., and Twilt, L., “Demonstrationof Real Fire Tests in Car Parks and High Buildings,”CTICM, Paris, France, 2002.

[14] Chen, M., Hu, C., and Chang, T. “The research on optimal parking space choice model in parkinglots,”In Computer Research and Development (ICCRD) 2011 3rd International Conference, March2011, pp 93-97.

[15] Wade, C., Baker, G., Frank, K., Robbins, A., Harrison, R., Spearpoint, M., Fleischmann, C.M.,“B-RISK user guide and technical manual,”SR 282. BRANZ Ltd, New Zealand, 2013.

[16] Stern-Gottfried, J., Rein, G., (2012) Travelling fires for structural design Part I: Literature review,Fire Safety Journal 54: 74-85, http://dx.doi.org/10.1016/j.firesaf.2012.06.003


957

http://dx.doi.org/10.1243/1748006XJRR169

http://dx.doi.org/10.1186/2193-0414-2-5

http://dx.doi.org/10.1007/s10694-006-0004-2

http://dx.doi.org/10.1016/j.firesaf.2012.06.003

Development of Fire Scenarios for Car Parking...

Documents

Transcript of Development of Fire Scenarios for Car Parking...